CloudTV × Codex

One prompt to bring Codex into your canvas

The CloudTV plugin bundles live MCP tools with reliable Skill workflows. Install it, authorize with OAuth, and use natural language to call app LLMs or read, plan, and edit your boards.

Open Infinite Canvas

OAuth protected8 text modelsModel choice cardsCredit result cards

Codex / cloudai

Natural prompt

CloudTV plugin

Infinite Canvas

MCP + SKILL

One connects your data. The other keeps work reliable.

MCP and Skill are installed together by the CloudTV plugin and load automatically in a new Codex task.

MCP: connect live boards

CloudTV OAuth gives Codex scoped access to your boards, live model catalog, and authorized text-model calls. It reads the current account instead of guessing.

OAuthcanvas.readcanvas.writellm.invoke

Skill: guide the workflow

Codex reads live models and boards. Text tasks use a model card before one paid LLM call; canvas mutations still require exact confirmation before an editable ChatCut handoff.

Read firstExact changesFresh approvalConflict safe

Install to create

Adapt Codex in four steps

Copy the install prompt

Paste the prompt below into a regular Codex task.

Install the cloudai plugin

Codex adds the marketplace and installs the Skill and MCP config.

Authorize with OAuth

Start a new task, sign in to cjai.pro, and approve the requested canvas or LLM scopes.

Describe your creative intent

Read and plan freely; confirm before Codex writes to the canvas.

One-prompt install

Give this prompt to Codex

Codex downloads and reviews the CloudTV installer, installs cloudai, and verifies its status. Start a new task after installation to load the Skill and MCP tools.

Expected plugin identifier

cloudai installed, enabled

The plugin bundles the CloudTV Infinite Canvas Skill, CloudTV LLM Skill, model and LLM-result cards, write-confirmation cards, and the https://www.cjai.pro/api/mcp connection.

install-cloudai.prompt

Install and configure the CloudTV Infinite Canvas plugin for me:
1. Download https://www.cjai.pro/downloads/install-cloudai.py to a temporary directory. Read it first and confirm it only installs to ~/plugins/cloudai, updates ~/.agents/plugins/marketplace.json, and calls Codex plugin commands.
2. After review, run python install-cloudai.py. It downloads the CloudTV bundle from www.cjai.pro and installs cloudai@personal.
3. Run codex plugin list and confirm cloudai shows installed, enabled.
4. Ask me to start a new Codex task so the plugin loads. In that task, open OAuth when prompted and sign in at https://www.cjai.pro.
5. After authorization, run “List my CloudTV canvas boards,” then “Use CloudTV LLM and let me choose the model in a card first.”
Use CloudTV interaction cards whenever I need to choose. Before every canvas write, present the exact mutation in a readable confirmation card and wait for my confirmation; do not print raw JSON.

Manual fallback+

Invoke-WebRequest https://www.cjai.pro/downloads/install-cloudai.py -OutFile "$env:TEMP\install-cloudai.py"
python "$env:TEMP\install-cloudai.py"
codex plugin list

Start creating

Operate the canvas like a conversation

Start with read operations. Choices use interactive cards; additions, deletions, and connections use expandable confirmation cards before approval.

Discover boards

“List my CloudTV canvas boards”

Read structure

“Read the nodes and edges in this board”

Plan and write

“Add a script node to the board and connect it”

Extend a workflow

“Add a shot script and connect it to the video node”

Choose models

“Use a card to let me choose script, image, and video models”

Send to ChatCut

“Prepare the board scripts and generated videos for an editable ChatCut project”

CLOUDTV LLM SKILL

Call every app text model from Codex with cards

cloudai-run-llm reads the live catalog before asking you to choose a model and reasoning level. One paid call runs only after card submission, followed by a result card with the answer, tokens, credits charged, and balance.

GPT-5.6 LunaRecommended

GPT-5.6 Terra

GPT-5.6 Sol

GPT-5.5

GPT-5.4

Gemini 3.1 Pro

Gemini 3.5 Flash

Gemini 3 Pro

Describe the text task

Provide the content to analyze, rewrite, summarize, translate, or generate.

Choose in a card

Codex reads the live catalog and shows model and reasoning options.

Submit one call

Submission authorizes one credit-consuming call with these settings.

Read the result card

See the answer, tokens, charged credits, balance, and request ID.

Copy this prompt to make a call

Use $cloudai-run-llm to turn the product requirement below into an implementation plan.
Read the current CloudTV text models first and show a card where I choose the model and reasoning level. Call the model only after I submit, then show the answer, token usage, credits charged, and remaining credits in a result card.

Requirement: replace this text with your task…

Reauthorize the llm.invoke scope after first install or a plugin update.
Calls use the current CloudTV credit balance; one submission authorizes one call.
The app live catalog is the source of truth; the Skill does not hard-code internal model IDs.
Uncertain network failures are not retried automatically to avoid duplicate charges.

Controlled writes

You keep the final say

Reads can run immediately. Board creation or node edits require a readable, non-JSON confirmation card. If the board changed, Codex re-reads and presents a fresh card instead of overwriting.

OAuth only accesses boards authorized by the signed-in account

canvas.read, canvas.write, and llm.invoke use separate scopes

Choices and write approvals use readable cards

One confirmation applies to one explicit write

Revision conflicts trigger a safe re-read

LLM and media generation never run automatically without confirmation

Ready to create with natural language?

With cloudai installed, Codex can call app LLMs through model cards, arrange script-to-video workflows, write approved changes, and prepare editable ChatCut handoffs.